“Did our trailer actually work?”
A film studio is about to spend millions putting a trailer in front of audiences. Before it does, it wants to know if the trailer makes people want to see the film, and if not, what to change.
Answering that is a campaign analytics study. This demo follows one study, start to finish, twice: the way it runs today, and the way it could run with a crew of AI agents.
Ember Falls: Trailer v3
Every day of process is a day less to fix the trailer.
A relay race: eight steps, seven hand-offs
This is the study's real path today, step by step. Watch the baton move: red steps are done by hand.
The research is fast. Everything around it is slow.
Respondents take the survey in minutes. The days are lost in the gaps between people and tools.
Copy, paste, re-key
A programmer hunts for an old study, clones its survey shell by hand and re-enters job details from the job system. The video is downloaded, watermarked and re-uploaded.
Chasing by email & Slack
Vendor specs go out by email. Quota progress is checked by hand in platform logs. A lagging audience group is often spotted late.
Cleaning after the fact
Fraud and junk answers are found only after fielding closes. SPSS scripts are hand-written per study, and charts are pasted into PowerPoint one by one.
Run the Ember Falls study both ways
Same brief, same 1,200 respondents, same 5 days in the field. We can't make people answer faster, so the agents shrink everything else.
⏳ Today: manual relayDay 0.0
⚡ With the agent crewDay 0.0
Four agents. Each one owns a stretch of the relay.
Pick an agent and watch one moment of its work on Ember Falls: what it notices, what it reads, what it decides and why, and what it hands back for a person to approve.
One connected platform, not eight separate tools
The agents work because the steps are joined up. The proposal uses two levers to remove every manual hand-off.
Where researchers and operators move data by hand today, connect the systems directly: job system ↔ survey platform, panel vendors, analytics store, report service.
Where no interface exists, an agent does the manual moves itself, driving the screens with browser automation (Selenium / Playwright) the way a person would.
Faster answers, cleaner data, and researchers back on insight
| Capability | Today | With the platform |
|---|---|---|
| Data processing turnaround | 3–5 business days | < 4 hours |
| Survey link setup & checks | 2–3 days | < 4 hours |
| Video watermarking | Manual download & re-upload | Automatic in the cloud |
| Fraud screening | After fielding, by hand | Live, during fielding |
| Fielding & vendor pacing | Spreadsheets & Slack | One live view, auto-balanced |
| Scoring vs. norms | File hand-off to data science | Automatic on clean data |
Campaign studies per year:
Placeholder until the savings calculation is final: assumes 8 days and about 38 team-hours saved per study.